Re: [PATCH] mtd: nand: omap: save Bad-Block-Table (BBT) on device

[Brian Norris <computersforpeace@gmail.com>]

Hi Ezequiel,

On Thu, Jul 24, 2014 at 10:40:18AM -0300, Ezequiel Garcia wrote:
> On 23 Jul 06:54 PM, Brian Norris wrote:
> > The only real backwards-compatibility concern you'd have for
> > unconditionally enabling on-flash BBT (like in this patch) is if you had
> > previous file system data in the last 4 blocks; nand_bbt will just
> > clobber it, breaking your file system. For this reason, using DT is
> > probably a good idea -- you're opting in, rather than being forced in by
> > a kernel upgrade.
> 
> FWIW, having the kernel wipe your precious data, is even worse than having
> him fail; which means this represents a much more serious drawback.
> This is why any modification to a NAND driver that involves a different
> 'view' of the flash, should be considered with a lot of care!

Yes, that's pretty serious. Perhaps there could be some more attention
paid to this issue (throwing out some half-baked ideas: tag partitions
as exclusive 'BBT' partitions, to ensure data / BBT separation; or
prevent nand_bbt from reserving non-empty blocks for BBT). But for now,
this issue can be best dealt with by using the supported DT option to
explicitly opt in.

> > Beyond the backwards-compatibility concern, you still have other
> > concerns about on-flash BBT's robustness. Limiting yourself to a region
> > of 4 blocks is one potential issue. There are others (e.g., lack of CRC
> > protection), but none that have been real show-stoppers. I have a few
> > generations of products running it here.
> 
> Hm.. Can you guys mention what's the benefit of using a BBT, against keeping
> the bad block mark in the OOB region? (Other than the fact that some ECC
> strength may not leave any available OOB).

The primary reason would be that NAND datasheets specify it these days
:) But a better argument: nobody guarantees that you can write a 
bad block marker to a worn out block; you may just get program failures.
This has been acknowledged by several developers over the last several
years, some of whom I think were quoting their recollection from talking
w/ manufacturers (does that make this 3rd-hand info?). Most recently,
Lothar brought this up [1].

Additionally, you get a boot-time performance improvement if you only
have to read a few pages, instead of a page or two from every block on
the flash.

Brian

[1] http://lists.infradead.org/pipermail/linux-mtd/2014-July/054767.html